An account team offers a bigger discount for a bigger, longer commitment, and it always looks like the same thing: free money on the table you would be foolish to leave there. Sign a three-year reservation instead of paying on demand and the unit price drops. Pre-commit to a spend floor and the whole bill gets cheaper. The math on the slide is airtight. The discount is real.
It is also a loan. You are borrowing against your own future flexibility, and the discount is the interest the lender pays you to make the loan. The cash you "save" this year is advanced against a promise to keep spending on the same vendor, the same architecture, and roughly the same shape, for as long as the term runs. The collateral is your right to change your mind. Most commitment decisions book the discount as income. It is the coupon on a loan whose principal is optionality, and the loan comes due whether or not the workload you committed to still exists.
I want to be honest about the seat I am arguing from. I do not sign the whole IT P&L. I run a real slice of it, the security, platform, and DevOps spend inside a fintech that serves more than 1,500 financial institutions, and I report its return to a board. From that slice the pattern is easy to name: the commitment gets sized to a forecast someone wanted to be true, booked as a saving, and then quietly outlived by the thing it was supposed to fund. This is not a case against committed spend. It is a case for pricing the loan before you take it.
The discount is the coupon. The collateral is your optionality.
Start with what actually changes hands. When you move from on-demand to a Savings Plan, a Reserved Instance, an Azure Reservation, or a multi-year Enterprise Agreement, the vendor lowers your unit price. In exchange you hand over a promise: a minimum you will spend, a term you will spend it over, and usually a payment structure that front-loads the cash. The vendor is not being generous. They are buying certainty. Your certainty is worth a discount to them because it lets them forecast revenue, plan capacity, and stop competing for the same dollar every quarter.
Notice who is really borrowing. The vendor advances you a discount today against future spend you have not made and budget you do not have. You repay it not in cash but in surrendered options: the option to shrink, to re-architect, to move volume to a cheaper tier, to walk to a competitor, to simply spend less next year because the business changed. That bundle of options has a value, and the deeper and longer the commit, the more of it you sign away. The discount is the price the vendor pays for it. The only question that matters is whether they paid you enough.
This is why the biggest advertised discount is so often the worst deal. The steepest reservation savings sit on the longest terms and the most-upfront payments, precisely the structures that surrender the most optionality. You are being paid the most to give up the most. Whether that is a good trade depends entirely on how much your optionality was worth, and that is the one number the slide never shows you, because only you can compute it.
Commit to the floor you'd run anyway, not the forecast you hope to hit
One asymmetry in almost every committed-spend contract should decide how you size it, and teams routinely size against it. You can nearly always spend above your commitment. The units past the floor bill at the undiscounted rate; you pay list on the margin, but nothing stops you. What you cannot do is spend below the commitment. Under-consume and you pay for capacity you never used, and the shortfall is money set on fire for nothing.
The two errors are not symmetric, so the sizing rule should not be either. Commit to the floor, the level of consumption you are confident you will run regardless of what happens to growth, the architecture, or the roadmap, and let everything above it float at on-demand rates. You give up some discount on the marginal units. In exchange you keep the very thing that discount was trying to sell you out of: the ability to be wrong about the future without paying a shortfall penalty for the privilege. Sizing to the optimistic forecast inverts this. It maximizes the paper discount and manufactures a stranded commitment the moment reality comes in under plan.
The vendor's forecast is not a neutral input. The account team is compensated to move you to the largest, longest commitment you will accept, and every planning conversation gets nudged toward the top of your range. Treat their number as an offer, not an analysis. The floor you will defend is the one you can prove from your own consumption history, not the one that makes this quarter's discount look best.
The interest rate on the loan is your own rate of change
Every loan has an interest rate, and on a committed-spend loan you set it yourself, by how fast you change. A workload that has been stable for two years on an architecture you have no plans to touch borrows cheaply: the flexibility you surrender is flexibility you were never going to use, so the commit costs you almost nothing beyond the paper. A workload that is actively migrating, with new instance families, a different database engine, a routing layer you are still tuning, borrows at a punishing rate, because the optionality you sign away is exactly the optionality you are about to need.
An internal agent platform like AgentOS is the clearest case of the type. That kind of platform rarely has a consumption profile that sits still. The moment you route work to a cheaper model or shift a service to a different compute family, a commitment sized to last quarter's shape is suddenly pointed at capacity you no longer buy. The commit does not get cheaper. Your rate of change goes up, and the loan reprices against you. The lesson generalizes past AI: the more legible and boring the workload, the longer you can safely commit; the more it is still moving, the shorter the term you can defend.
Lock-in compounds the rate. A deep multi-year commit lowers your unit price and raises your switching cost, because leaving the vendor now means eating the stranded remainder of a commitment you already paid for. The discount quietly becomes a moat, and it is the vendor's moat, not yours. In a regulated shop that is not only a cost concern; concentration and continuity of a critical vendor are exactly the questions an examiner asks, and "we cannot move because we pre-paid three years" is not an answer that inspires confidence in the room. The flexibility premium is real. Pay it with your eyes open, on the workloads that have earned a long commit by holding still.
Right-size the commit across the portfolio, not one contract at a time
The mistake that turns a manageable set of loans into a balance-sheet problem is taking each one in isolation. The cloud Enterprise Agreement gets negotiated by the platform team. The data-warehouse capacity commitment gets signed by data engineering. The observability multi-year gets renewed by whoever owns the tool. Each looks reasonable alone. Stacked, they quietly consume a share of next year's budget that nobody ever decided to pre-spend: a pile of correlated obligations, all coming due on their own schedules, all shrinking the discretionary dollars you have left for anything you did not foresee.
Committed spend deserves a portfolio limit the way leverage deserves one. Set a ceiling on total committed spend as a share of total addressable spend, and hold the whole vendor book under it rather than each contract. The exact number is a judgment call, more conservative the faster the business changes. The discipline is what matters: a single owner who can see every commitment at once and say no to the next one because the aggregate is already at the line. Before you sign anything, you should be able to answer four questions about the whole book.
- What share of next year's spend is already committed across every vendor (cloud, data platform, observability, security tooling, licensing) before a single new decision is made.
- What the coverage and utilization actually are on the commitments you already hold. The FinOps Foundation's commitment metrics, coverage, utilization, and effective savings rate, exist precisely so this stops being a feeling. Low utilization means you are already paying a shortfall you never named.
- How correlated the obligations are. Commitments that all assume the same growth story fail together the year that story does not land.
- How much discretionary spend survives the committed floor, because that residual is your actual capacity to absorb surprise, and every new commit spends it down.
Committed spend is a liability the P&L is happy to hide
This discipline is hard to hold because the accounting flatters the decision. The discount shows up immediately and legibly as a lower run rate, a number you can announce. The obligation shows up nowhere the quarterly conversation looks. I am a technology leader reading the shape of this, not your controller, so take the specifics to someone who signs the statements. As I understand it, a multi-year purchase commitment is a contractual obligation that typically lives in the footnotes of the financials as future minimum payments: disclosed, but not sitting on the face of the income statement as a debt you feel. The savings get celebrated on the P&L. The liability gets filed where nobody managing the roadmap is reading.
So report the other half yourself. When committed spend is booked as pure savings, the person approving it sees the coupon and none of the principal. The honest report shows both: the discount we captured, and the forward obligation and optionality we spent to get it. How much of next year is now pre-committed, what our coverage and utilization are running, and what it would cost to exit early if the workload moved. That is the version a board or a budget owner can actually decide against, because it prices the loan instead of just booking the interest.
Size the loan, don't just take the discount
Commit to the floor you would run anyway, never the forecast you are hoping for. Set the term by your rate of change, not by the size of the advertised discount. Cap total committed spend across the whole portfolio, and give one owner the authority to say the book is full. And report the forward obligation next to the saving, so the people funding it are pricing the loan instead of pocketing the coupon.
The commit you will regret is never the one that looked expensive. It is the one that looked free: sized to a forecast nobody had to defend, on a workload that moved six months later. Price the optionality before you sell it, or you will find out what it was worth at the renewal, when the vendor already knows.